This year on the EGU General Assembly blog there will be guest posts from participants about their research and their impressions of sessions. These are personal points of view not EGU corporate views. If you would like to contribute a research viewpoint, please email us.
This post comes from Aidan Slingsby of City University, London, who looks at how we make sense of large datasets using visualisations .
My research is about designing visualisation methods to help make sense of large datasets. I’m based at the giCentre at City University London and our group’s research focuses on design, implementation, user-engagement and user-evaluation. We primarily deal with data with a geographical component, but not exclusively so. I’m part of the Willis Research Network, through which I apply these techniques to some of the needs of the insurance industry.
I believe that well-designed, fast and responsive visual interfaces to data have an important role, particularly in the early stages of data analysis. Such interfaces increasingly incorporate some (though usually limited) analytical capabilities, such as comparisons to models of expectation. Providing the flexibility to query large datasets on-demand to pursue research questions is conducive to insight discovery. Hypotheses generated through this initial data exploration can be subsequently verified.
Key to success here is appropriate design. This is tricky. There are many facets to design, many of which are context dependent – the nature of the data, the experience of the users, the users’ level of engagement with the data. Some designs have widespread appeal, but do not offer much new insight into data. Some designs offer sophisticated comparison and analytical capability, but are impossible for the target users to use. Some designs simple don’t show that aspects of the data that the user is interested in. With so many factors that affect whether designs are “good”, designing usually is not straightforward.
Our group designs for a range of users and research questions and our membership of the Willis Research Network helps provide a context within which we can work.
This year, we are reporting our work in two posters. The first “Browsing large natural hazard event sets” (NH9.1/EG8) is work with the National Centre for Atmospheric Science at Reading University here we are using interactive visualisation to browse through 150 years of simulated storms and the associated atmospheric conditions. We taking a user-centred approach, trying to produce designs that can help allow such a large dataset to be interactively browsed and help answer specific types of research question. We’ll be in attendance at the poster between 1730 and 1900 on Thursday and will be pleased to show you this work. This is based on some work we did last year that is written up here.
The other poster “Sharing insights on the impact of natural disasters on Twitter” (EGU2011-9171) is work that tries out an idea of ours – can we better engage people with datasets by giving them the visualisation tools they need to produce useful views of the data and then share these insights with others. To try this idea out, we provided users of earthquake-report.com a copy of some software called HiDE and a dataset (that we knew they were interested in) of some of the impacts of natural catastrophes over the past 30 years worldwide. We them asked them to construct some graphics that told them something meaningful about the data and encouraged them to share it through Twitter using our software. Details are here.
This experiment was not a success, in the sense that there was very low participation (although many page views) and no one completed our questionnaire about their experiences and thoughts. Although it was rather a speculative experiment and it would have difficult to draw too many conclusions from what we found, this evidently did not capture people’s imagination in the way we’d hope. There are many possible reasons for this, including many of the points I’ve already made. It may have been too difficult to use, the graphics may have not been unhelpful, people may not have been interested in the dataset enough to want to spend time exploring it, people may have been too busy to take part, people may have considered it not a good use of their time, people may not have had Twitter accounts. Some/many aspects of the design of this experiment was/were wrong. Feel free to have a go yourselves and let me know how you get on!
I’m happy to hear your views of anything I’ve said – just find me at EGU on Thursday or email me.